MCP for AI agents

MCP gives an external AI agent working access to the platform: to read analytics, help with surveys, and perform actions on behalf of the user. The connection goes through a Streamable HTTP endpoint:

https://<platform-domain>/mcp

On the first request, the agent opens an OAuth login in the browser. You see which client is requesting access and what scopes it requests. After confirmation, the token is stored by the MCP client itself; you do not need to copy the secret manually.

One endpoint works for all your organizations. The active organization is taken from the authorized session; if the agent supports organization switching, it will be able to select another available organization.

Claude

For Claude Code, add an HTTP MCP server:

claude mcp add --transport http tealband https://<platform-domain>/mcp

Open Claude Code and run /mcp, then go through OAuth in the browser. If you need to trigger the login from the terminal, use:

claude mcp login tealband

Codex

Add a Streamable HTTP server:

codex mcp add tealband --url https://<platform-domain>/mcp

Then perform OAuth login:

codex mcp login tealband

In the Codex CLI, the connection status is visible via /mcp.

OpenCode

Add the server to opencode.json:

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "tealband": {
      "type": "remote",
      "url": "https://<platform-domain>/mcp",
      "enabled": true
    }
  }
}

On first use, OpenCode will prompt to authorize. You can manually launch OAuth like this:

opencode mcp auth tealband

OpenClaw

Save the MCP server as Streamable HTTP and enable OAuth:

openclaw mcp add tealband \
  --url https://<platform-domain>/mcp \
  --transport streamable-http \
  --auth oauth

Start the login and test the connection:

openclaw mcp login tealband
openclaw mcp doctor tealband --probe

If login returns a confirmation code, complete the login with the command:

openclaw mcp login tealband --code <code>

If editing the config manually, the format is as follows:

{
  "mcp": {
    "servers": {
      "tealband": {
        "url": "https://<platform-domain>/mcp",
        "transport": "streamable-http",
        "auth": "oauth"
      }
    }
  }
}

Hermes

Add the server to ~/.hermes/config.yaml:

mcp_servers:
  tealband:
    url: "https://<platform-domain>/mcp"
    auth: oauth

Start Hermes. On the first connection, it will open OAuth in the browser and save the tokens for subsequent sessions. After changing the config in an already open session, execute:

/reload-mcp

Access

The agent only receives confirmed scopes:

• analytics:read — read surveys, results, analytics, and growth points;
• surveys:write — create, edit, and publish surveys;
• org:manage — manage organizations, settings, and participants.

Scopes do not replace the user's role. If your role does not allow you to perform an action in the organization, the agent will not be able to perform it either.

Privacy

MCP does not open participant responses bypassing the platform's rules. No one sees raw answers; anonymized data and aggregates are shown externally only with a sufficient number of answers (at least 5). If there is little data, the agent will receive a limited result.

For regular scripts and server integrations, use API tokens. MCP is specifically needed for AI agents that require tools, resources, prompts, OAuth, and individual scopes.